I have a set of fewer than 1 000 complete fingerprint images and I want to teach a convolutional neural network, in MATLAB, to rebuild their missing regions. The workflow starts with generating partial samples from each full print using my own predefined masks; no random cropping or synthetic noise, just the masks I supply. Here’s what I expect the finished project to cover: • A script that reads the full prints, overlays every predefined mask and saves the visible fragments as training inputs while keeping the original images as ground-truth targets. • A CNN architecture implemented with MATLAB’s Deep Learning Toolbox (layers, training options, data augmentation, checkpointing). Feel free to suggest a proven structure such as U-Net or an encoder-decoder, as long as it runs end-to-end in MATLAB. • Training code that accommodates the modest data size, applies validation splits, and prevents over-fitting (transfer learning or regularisation are welcome). • Inference code that takes an unseen partial, feeds it to the trained model, and outputs the reconstructed fingerprint. • An evaluation routine reporting reconstruction quality (e.g., SSIM, PSNR) and optionally a side-by-side montage for quick visual checks. Everything should be neatly commented and ready to run on a standard MATLAB install with the Deep Learning Toolbox. If you rely on custom toolboxes or external functions, include them or point me to the exact files. Once the scripts reproduce the reconstruction results on my machine and hit the agreed metrics, the job is done.